JP2013196653A - Interest analysis method, device and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable a user to obtain a relatively good recommendation result even in the case of using an information recommendation service for the first time.SOLUTION: A model of each attribute of the existing information user is created and stored on the basis of information representing a content selection history by the information user. Then, attribute information of a new information user is acquired, a model of an attribute of the existing information user coincident with an attribute of the new information user is selected from in the stored model group on the basis of the acquired attribute information, a model expressing the degree of interest in a prescribed concept of the new information user is created on the basis of the selected model, and the interest of the new information user is analyzed on the basis of the created model.

Description

  The present invention relates to an interest analysis method, apparatus, and program for analyzing a user's interest from a content browsing history or the like.

  There is a demand for a technique for recommending appropriate services and contents in accordance with user behavior and situations. In view of this, an information recommendation service has been proposed in which information suitable for each information user is selected or information is rearranged in an order suitable for each information user and displayed. For example, when the information is an advertisement, it is an advertisement recommendation service that displays an advertisement by selecting an advertisement suitable for each information user or rearranging in an order suitable for each information user, and when the information is news It is a news recommendation service that selects news that is appropriate for each information user, or displays the news in an order that is appropriate for each information user. If the information is restaurant information, it is appropriate for each information user. It is a restaurant information recommendation service that displays restaurant information by selecting restaurant information or rearranging in order suitable for each information user.

  As a method for realizing the above information recommendation service, the user's interest is estimated from the frequency of the concept or the like appearing in the user history on the premise that the concept tag (metadata) for summarizing the contents is assigned to each content. There is a method, and this method is a content-based filtering method (Content Based Filtering: CBF).

  Specifically, the content-based filtering technique presents products of the same brand (holding the same concept tag) when browsing products of a specific brand (holding information indicating the brand as a concept tag). In this case, the memory-based method presents a product of a specific brand if the specific brand is frequently browsed from the history of browsing in the past. In a simple method, products related to concept tags that appear more frequently in the browsing history are presented. As a method using a taxonomy (ontology) in such a method, for example, there is “Proposal of a multifaceted preference grasping method using an operation history” in Non-Patent Document 1.

"Proposal of multifaceted preference grasp method using operation history", IEICE Technical Report, vol. 110, no. 450, LOIS2010-76, pp. 67-72, March 2011

  However, the technique described in Non-Patent Document 1 has the following problems to be solved. That is, when the information user uses the information recommendation service for the first time, or when the information user has rarely used it, there is no or very little user history. For this reason, the user history necessary for creating the model is insufficient and an appropriate model cannot be created, and as a result, appropriate recommendation is not performed. This is called a Cold Start problem and causes the following problems. That is, when using the information recommendation service, since the user history is small and appropriate recommendation is not performed, the satisfaction of the information user is hardly increased. Therefore, the information recommendation service is not used and the user history is not accumulated. As a result, since the user history is small, an appropriate recommendation is not performed, and a negative chain occurs. As a result, the number of information recommendation service users who stop using the service increases, and the information recommendation service does not spread.

  This invention was made paying attention to the above circumstances, and its purpose is an interest analysis method capable of obtaining a relatively good recommendation result even when the user uses an information recommendation service for the first time, To provide an apparatus and a program.

In order to achieve the above object, one aspect of the present invention is that a computer estimates information user's interest using information representing the scarcity of concept appearance and uses the estimation result for each information user. In a method for creating a model representing the degree of interest in a predetermined concept and analyzing the interest of each information user using the created model, information representing a content selection history by the information user is obtained. Originally, a model for each attribute of the existing information user is created and stored in the first storage unit. Then, the attribute information of the new information user is acquired, and the model of the attribute of the existing information user that matches the attribute of the new information user based on the acquired attribute information is stored in the first memory. A model representing the degree of interest in the predetermined concept of the new information user is created based on the read model and selectively stored in the second storage unit as an initial model. save. Then, the interest analysis of the new information user is started using the model stored in the second storage unit as an initial model.
Therefore, even if the user uses the information recommendation service for the first time, a relatively good recommendation result can be obtained.

One aspect of the present invention is characterized by comprising the following aspects.
In the first aspect, when the model for each attribute of the existing information user is created, the attribute of the existing information user has a first hierarchy corresponding to a lower concept and a higher hierarchy than the first hierarchy. A second-level model corresponding to the concept, a lower-level concept model corresponding to the first level is created, and the second-level model is created based on the lower-level concept model created for the first level. The model of the superordinate concept corresponding to the hierarchy of is created.
In this way, when the attribute has a hierarchical structure, a model of the higher-level concept attribute is created using the model of the lower-level concept attribute. Is possible.

In the second aspect, a threshold value or a time zone in which the processing load of the computer is equal to or less than a preset value is set, and when the processing load is equal to or less than the set threshold value, or the preset time zone has been reached In some cases, processing for creating a model for each attribute is executed.
In this way, the process of creating a model for each attribute of an existing information user is performed by selecting a midnight time zone where the processing load on the computer is low, etc., so it does not increase the processing load on the computer. can do.

  That is, according to the present invention, it is possible to provide an interest analysis method, apparatus, and program capable of obtaining a relatively good recommendation result even when a user uses an information recommendation service for the first time.

1 is an overall configuration diagram of a system using an interest analysis device according to an embodiment of the present invention. The block diagram which shows the function structure of the client terminal shown in FIG. 1, a content server, and an interest analysis apparatus. The figure which shows the outline | summary of a process of the interest analysis apparatus shown in FIG. The figure which shows an example of content browsing operation on the client terminal shown in FIG. The figure which shows the data structural example of a list browsing content list. The figure which shows the data structural example of detailed browsing content. The figure which shows the data structural example of a presentation content list. The figure which shows an example of a content database. The figure which shows an example of a concept system / user interest score database. The figure which shows the processing flow of a log | history information receiving part. The figure which shows the processing flow of a characteristic score calculation part. The figure which shows the data structural example of an analysis parameter list. The schematic diagram for demonstrating operation | movement of a characteristic score calculation part. The figure which shows the detail of a characteristic score calculation process. The figure which shows the processing flow of a concept system update process part. The figure which shows the detail of a concept system update process. The figure which shows the processing flow of a content evaluation process part. The figure which shows an example of a content score list. The figure which shows the detail of a content evaluation process. The figure which shows an example of content request data. The figure which shows the 1st example of the attribute model creation process by the interest analysis apparatus shown in FIG. The figure which shows the 2nd example of the attribute model creation process by the interest analysis apparatus shown in FIG. The figure which shows the example of the initial model creation process by the interest analysis apparatus shown in FIG. Schematic used for explanation of the principle of the present invention. The figure for demonstrating the model creation process for every attribute. The figure which shows the 1st example of the update formula of a user interest score. The figure which shows the 2nd example of the update formula of a user interest score. The figure for demonstrating the outline | summary of the initial model creation process using the model for every attribute. The figure which shows an example of the calculation formula used for the initial model creation process using the model for every attribute. The figure for demonstrating the congestion of a model update process. The figure used for description of the model update process for every attribute using a user's model. The figure which shows an example of the calculation formula used for the model update process for every attribute using a user's model. The figure used for description of model creation processing for every attribute with a hierarchical structure using a user model. The figure which shows an example of the calculation formula used for the model creation process for every attribute with a hierarchical structure using a user model.

[principle]
First, the principle of the present invention will be described with reference to FIGS.
This invention estimates the interest of an information user with high precision using the scarcity of concept appearance, creates a model for each information user using the estimation result, and uses this created model to create information Assuming a method of recommending In this premise method, when the information user estimates the information user's interest from the history of selecting the information from the selection candidate information list, an information list that is a selection candidate of the information user is defined, and from there By analyzing the number of occurrences of concepts in information selection, the rareness of each concept appears. Furthermore, the rarity where each concept does not appear is also used. For this reason, it becomes possible to estimate an interest of an information user with high accuracy. Hereinafter, the selection candidate information list presented to the information user and the history in which the information user has selected arbitrary information from the selection candidate information list are referred to as an information selection history.

The present invention further improves the above-described method so that a relatively good recommendation result can be obtained even when the user uses an information recommendation service for the first time.
FIG. 24 is a diagram for explaining the outline of the improved invention. As shown in FIG. 24, the interest analysis apparatus according to the present invention executes the following process to recommend highly reliable information to the new user F.

(1) Using the history of existing users (for example, users A to E in FIG. 24), for each attribute, for example in the example shown in FIG. 24, for each residence such as Tokyo and Chiba, and for each sex such as male and female Create
(2) A part of attribute information that the new user F permits to disclose is disclosed to the system.
(3) A model that matches the attribute disclosed by the new user F, for example, the Tokyo model and the female model are selected in the example of FIG.
(4) Create an initial model for the new user F using the selected model.
(5) The new user F starts using the service using the created initial model, and then continues using it. As a result, the model of the new user F is gradually corrected from the initial model representing the stereotype of Tokyo and women to the model representing the new user F itself, and the recommendation accuracy is improved.

  By executing such processing, the following effects can be obtained according to the interest analysis apparatus according to the present invention. That is, by using an appropriate initial model, relatively good recommendation accuracy is realized from the start of use, and high satisfaction is realized, thereby promoting the use of the information recommendation service. For this reason, accumulation of the information selection history proceeds, the degree of adaptation to the user's own interest increases, and the recommendation accuracy is further improved. As a result, the number of information recommendation service users who stop using the service is reduced, and the information recommendation service becomes widespread.

Next, a method for creating a model for each attribute using the history of existing users in the series of processes shown in FIG. 24 will be described with reference to FIG.
In one learning, the user's model is updated using a history of browsing a plurality of contents as a list and a history of browsing the content body. Here, the model for each user is a set of user interest scores (TotalZ) calculated for each concept. When obtaining the user interest score (TotalZ) for each concept, first cluster a first content list in which a plurality of contents are browsed as a list and a second content list in which content bodies are browsed from the first content list. Turn into. This clustered content list is called a history cluster.

  Next, for each history cluster, the total number of contents in the first content list is the first total number, the number of contents in which the concept appears in the first content list is the first appearance number, and the second When the total number of contents in the content list is expressed as a second total number and the number of contents in which the concept appears in the second content list is expressed as a second appearance number, these first total number, first A first probability that the number of contents in which the concept appears in the second content list is greater than or equal to the second occurrence number under the condition of the occurrence number and the second total number, and less than or equal to the second occurrence number The second probability is calculated.

  Based on the first and second probabilities, a feature score (Z) is calculated by an inverse function of the cumulative distribution function of the standard normal distribution, and the feature score calculated for each history cluster is used to calculate the feature score. Calculate and update user interest scores. FIG. 26 shows an equation for calculating the user interest score using the feature score. Here, the weight W is a value set for each concept in each history cluster.

  Next, as shown in FIG. 25, the model for each attribute is updated simultaneously with the updating of the user's model. For example, when updating a male model, when a history cluster of a user whose attribute is “male” (user A, user B, user E in the example shown in FIG. 25) is obtained, the model of user A, the user At the same time as updating the model of B and the model of user E, the male model is also updated. FIG. 27 shows a formula for updating the interest score (TotalZ) for each concept i of the attribute C using the history cluster of the user A that matches the attribute C.

  Subsequently, an initial model using a model for each attribute is created. FIG. 28 is a diagram for explaining the creation method. For example, when the new user F presents the attributes “Tokyo” and “female” to the information recommendation system, the information recommendation system shows the Tokyo model and female model that match the attributes of the new user F in FIG. Combined using equations, used as an initial model for new user F.

  However, if a method for updating a model for each attribute each time a history cluster is obtained as shown in FIG. 25, for example, a model with a high ratio to the number of users such as a male model or a female model is updated. In this case, the update frequency of the model increases as the number of service users increases. As a result, the load on the system increases, which may cause delays in update processing, omission of updates, and system shutdown (FIG. 30).

  This problem can be solved as follows. FIG. 31 shows a proposal obtained by further improving the method for updating the model for each attribute shown in FIG. The model update for each attribute is performed using the learning result for each user instead of using the history cluster of each user that matches the attribute. However, since the number of learnings is different for each user, if the interest scores of the users are stochastically combined at the same rate, the difference in the number of learnings is not considered. Accordingly, X and Y stored together are used instead of the interest score of each user. X and Y depend on the number of times each user learns. For this reason, the difference of the learning frequency for every user can also be considered by considering these.

When Yi and Xi are determined for all users related to attribute C, the nth interest score (TotalZ _Cin ) related to the concept i of the model related to attribute C and the formula for updating Y _Cin and X _Cin are shown in the figure. 32.
Further, FIG. 33 shows an example of hierarchically creating a model for each attribute. For example, when a Kanto model has a Tokyo model and a Chiba model as children, as shown in FIG. 34, the X _{child attribute i} and Y _{child attribute i} of the _{child attribute} model are used, and the X _{parent of the} parent attribute model is used. _{Attribute i} , Y _{parent attribute i} , and interest score (TotalZ _{parent attribute in} ) can be obtained.

[Embodiment]
Hereinafter, an embodiment according to the present invention will be described with reference to the drawings.
(1) Overall Configuration of System FIG. 1 is an overall configuration diagram of a system using an interest analysis apparatus according to an embodiment of the present invention. This system includes a client terminal 200, a content server 300, and an interest analysis device 100. The client terminal 200 and the content server 300, and the content server 300 and the interest analysis device 100 are connected to each other via a communication network. The user acquires desired content from the content server 300 through a browsing operation on the client terminal 200, displays the acquired content on the screen of the client terminal 200, and browses.

  The client terminal 200 collects a content browsing history by a user operation, browses a plurality of content as a list, a list browsing content list (first content list), and a detailed browsing content list that browses the content body from the content list (Second content list) is transmitted to the content server 300. The content server 300 transfers the list browsing content list and the detailed browsing content list to the interest analysis device 100 via the communication network.

The interest analysis apparatus 100 calculates a feature score and a user interest score for each concept appearing in the content based on the list browsing content list and the detailed browsing content list, and estimates the user's interest. Based on this user interest score, the interest analysis apparatus 100 generates a list of contents sorted according to the user's interest (sorted presented content list) from the “presentation content list” received from the content server 300. To the content server 300.
In addition, the interest analysis apparatus 100 periodically creates an interest model of an attribute using a user's interest model, and creates an initial model according to the attribute when receiving attribute information from an initial use user. Also equipped.

FIG. 2 is a block diagram illustrating functional configurations of the client terminal 200, the content server 300, and the interest analysis apparatus 100 illustrated in FIG.
(2) Configuration of Interest Analysis Device The interest analysis device 100 includes a history information reception unit 110, a feature score calculation unit 120, a concept system update processing unit 130, a concept system / user interest score database 140, a presentation content list reception unit 150, content A database 160, a content evaluation processing unit 170, and a sorted content score list transmission unit 180; an attribute interest score update activation unit 190; an attribute interest score update unit 195; an attribute interest score database 196; and an initial user interest score creation unit 197. And a user / attribute information database 198.

FIG. 3 shows an outline of processing of the interest analysis apparatus 100.
The history information receiving unit 110 receives the list browsing content list and the detailed browsing content list from the client terminal 200 via the content server 300. The list browsing content list is, for example, a list of content in which the user browses only the content titles in a list. The detailed browsing content list is a list of content that the user has viewed the content (details) of the content body. For example, in FIG. 3, the list browsing content list includes contents 1 to 8, and the detailed browsing content list includes contents 1, 3, and 4. In FIG. 3, the content indicated by the hatched pattern indicates that the concept B appears in the contents 1, 6, 7, and 8.

The feature score calculation unit 120 calculates a feature score (Z value to be described later) of each concept by a statistical model of concept selection using the list browsing content list and the detailed browsing content list.
The concept system update processing unit 130 updates the user interest score for each concept based on the relationship information (superordinate concept and subordinate concept) in the concept system using the feature score. Nodes included in the graph of the concept system represent concepts, and links represent relationships between concepts. The user interest score is held as a value of a node corresponding to each concept in the concept system. In the concept system, the nodes located at the higher level represent the abstract concept, and the nodes located at the lower level represent the specific concept. The concept system and concept ID (identifier assigned to each node) are designed and defined in advance by a service operator or the like.

  The content evaluation processing unit 170 calculates the user's evaluation score for the content by probability combining using the user interest score of each concept appearing in the evaluation content. In the example of FIG. 3, the evaluation score of the content 1 is obtained using the user interest scores of the concepts E, F, and D that appear in the content 1. Detailed operation description of the interest analysis apparatus 100 will be described later.

The attribute interest score update activation unit 190 periodically activates an attribute interest score update process in a time zone with a low load such as a midnight time zone.
The attribute interest score database 196 is used to store X, Y, and TotalZ of attributes calculated by the attribute interest score update unit 195 described later.
The user / attribute information database 198 is used for storing user information and its attribute information.

  The attribute interest score update unit 195 uses the attribute information stored in the user / attribute information database 198 to specify a user group that matches the attribute for each attribute, and based on the X and Y of the specified user. 32 has a processing function for obtaining X, Y, and TotalZ of the attributes using the formula shown in FIG. 32 and storing the obtained results in the attribute interest score database.

  Also, the attribute interest score update unit 195 specifies attribute child attribute groups using attribute information stored in the user / attribute information database 198, and the attribute interest score update unit 195 is shown in FIG. 34 based on the X and Y of the specified attributes. The processing function for obtaining X, Y, and TotalZ of the above-described attributes using the formulas and storing the obtained results in the attribute interest score database 196 is also provided. X, Y, and TotalZ will be described later.

  The initial user interest score creation unit 197 reads an attribute model that matches the attributes of the new user transferred from the client terminal 100 via the content server 300 from the attribute interest score database 196. Then, the read attribute model is stochastically coupled using the formula shown in FIG. 29, and the result is stored in the conceptual system / user interest score database 140 as an initial model of the initial use user.

(3) Configuration of Client Terminal In FIG. 2, the client terminal 200 includes a history collection unit 210, a history information transmission unit 220, a content presentation unit 230, a content request transmission unit 240, and an attribute presentation unit 250.
The content request transmission unit 240 makes a content presentation request to the content server 300 in accordance with a user instruction (input). Specifically, content request data as shown in FIG. 20 is transmitted to the content server 300. For example, the content request data has a client terminal ID (or user ID) and a request time. The request time may be added in the content server 300. The client terminal ID (or user ID) is a number uniquely assigned to each terminal (or user), and is an ID that matches the user ID of the user interest score table of the conceptual system / user interest score database 140 described later. is there.

FIG. 4 shows an example of a content browsing operation by the user on the client terminal 200.
Based on the sorted presentation content list received from the content server 300, the content presentation unit 230 displays a list from the top of the sort order within the range allowed by the display screen size of the client terminal 200.

  In the example shown in FIG. 4, ten contents (contents 1 to 10) are displayed in a list. The content in the lower order of the sort order can be displayed by the list by the user's flick, scroll bar operation or the like. The list of contents actually displayed on the client terminal 200 in this way is referred to as a list browsing content list. That is, not all the contents in the sorted presentation content list are displayed on the client terminal 200, and thus are not necessarily included in the list browsing content list. When the user selects a title of each content from this list by clicking or the like, the main body (details) of the content of the selected title (contents 3, 5, and 6 in FIG. 4) can be viewed. The content whose details are browsed is included in the detailed browsing content list.

  As described above, the history collection unit 210 collects user operation histories and creates a list browsing content list and a detailed browsing content list. The history information transmission unit 220 transmits the list browsing content list and the detailed browsing content list created by the history collection unit 210 to the content server 300.

  FIG. 5 shows a data configuration example of the list browsing content list in the case of FIG. The list browsing content list has a cluster ID, a content ID, and a browsing time. The cluster is an identifier ("1" in FIG. 5) uniquely assigned to the list browsing content list and the detailed browsing content list. When the user browses the list browsing content displayed at another time (time zone), another cluster ID is given. The conditions for newly assigning the cluster ID under conditions other than the time include cases where there is no operation for a certain period of time while the list browsing content list is displayed, when the location is moved, when the temperature or weather changes, and the accompanying person May change, the browsing user (user ID) may be switched, the narrowed search may be applied to the list browsing content list from the viewpoint of the content genre, or the browsing mode may be switched in other browsing applications. The content ID is an identifier uniquely assigned to each content of the list browsing content, and is assumed to match a value held in a content database 160 described later.

  FIG. 6 shows a data configuration example of the detailed browsing content list in the case of FIG. The detailed browsing content list has a cluster ID, a content ID, and a browsing time, like the list browsing content list. The cluster ID is set to the same value as the list browsing content list (“1” in FIG. 6). In the detailed browsing content list, the content ID and browsing time are the identifier of the content (contents 3, 5, and 6 in FIG. 6) that the user has selected from the browsing content and browsed the content and the time when the content was browsed.

  The attribute presenting unit 250 has a function of accepting input of this attribute information and transmitting it to the content server 300 when a user who uses this information recommendation service for the first time, that is, a new user (initial use user) inputs his / her attribute information. Have

(4) Configuration of Content Server In FIG. 2, the content server 300 includes a content transmission processing unit 310, a sorted presentation content list reception unit 320, a presentation content list transmission unit 330, a presentation content list input unit 340, and a history information transfer unit. 350, a content request transfer unit 360 and an attribute transfer unit 370.

The history information transfer unit 350 transfers the list browsing content list and the detailed browsing content list received from the client terminal 200 to the interest analysis device 100 via the communication network.
The presentation content list input unit 340 receives a presentation content list that lists contents to be presented to the client terminal 200 used by the user by the service operator. The presented content list transmission unit 330 transmits the input presented content list to the interest analysis apparatus 100 via the communication network.

  FIG. 7 shows a data configuration example of the presented content list. The presented content list has a content ID, a concept ID / relevance list, a content body, and a content registration time. The content ID is a unique ID assigned by the content server 300 to each content. The concept ID / relationship degree list stores a concept ID of a concept that appears in content and a set of values indicating the degree of relevance between the concept and the content. The concept ID / relevance degree list is set in advance for each content. As a specific example, content 1 (sports article) includes {“baseball” concept ID = 1, relevance = 0.5}, { A set of concept ID and degree of association is stored as “soccer” concept ID = 2, degree of association = 0.8}, {“golf” concept ID = 3, degree of association = 0.6}. .

  The concept ID matches the value stored in the concept system / user interest score database 140. For example, the relevance is a value from 0 to 1, and the larger the relevance, the stronger the relevance. As the relevance, a value set by the service operator at the time of content registration or a value calculated by another system is used.

  The sorted presentation content list receiving unit 320 receives a sorted presentation content list and a client terminal ID (or user ID) obtained by sorting a part or all of the presentation content list from the interest analysis device 100. The content transmission processing unit 310 transmits the sorted presentation content list to the client terminal 200 corresponding to the client terminal ID (or user ID).

The content request transfer unit 360 transfers content request data (FIG. 20), which is a content presentation request from the content request transmission unit 240 of the client terminal 200, to the interest analysis device 100.
The attribute transfer unit 370 has a function of receiving the new user attribute information transmitted from the client terminal 200 and transferring the received new user attribute information to the interest analysis apparatus 100.

(5) Operation of Interest Analysis Device Next, the operation of each part of the interest analysis device 100 will be described.
(5-1) Content database 160
FIG. 8 shows an example of the data structure of the content database 160. The content database 160 has a content table and a user history table.
The content table stores a content ID, a concept ID / relevance list, a content body, and a content registration time. The value received by the presented content list receiving unit 150 is stored.

  The user history table stores content ID, user ID (client terminal ID), detailed browsing total number, detailed browsing time, list browsing total number, list browsing time, and list non-display flag. The detailed browsing time is null when the total number of detailed browsing is 0, and if it is 1 or more, a list of browsing times in a time series of each browsing is stored. The list browsing time is null when the total number of browsing the list is 0, and if it is 1 or more, a list of browsing times according to the time series of each browsing is stored. The list non-display flag stores false when the user has not yet displayed / viewed as a list on the screen of the client terminal, and stores true when the list has been viewed even once. In the user history table, all content ID values are held for each user ID. As the detailed browsing total number and the list browsing total number, when a large number of list browsing content lists indicated by the cluster ID are received, a cumulative total of past histories is stored.

  For example, by using the data of the user history table, the evaluation score of the content is lowered in the future evaluation (during processing in the content evaluation processing unit 170) according to the number of times the user browses. As a method for reducing the evaluation score, when the number of browsing for a certain content is k, the evaluation score of the content is divided by k + 1, or the evaluation score is multiplied by a weight (for example, 0.9) to the kth power. By this process, it is possible to reduce the repeated presentation of the same content in consideration of the degree of coincidence with the interest, so that the satisfaction with the user's recommendation can be improved.

(5-2) Conceptual system / user interest score database 140
FIG. 9 shows an example of the data structure of the conceptual system / user interest score database 140. The concept system / user interest score 160 includes a root concept node ID, a concept system table, and a user interest score table.
The root concept node ID is the highest concept node ID in the concept system structure, and only one root node exists in the system.

The concept system table stores a self-concept ID, a parent concept ID list, and a child concept ID list. All self-concept IDs in the concept system are stored in association with a parent concept ID and a child concept ID (however, if the self-concept is the lowest, there is no child concept ID), thereby defining the concept structure Is done.
The user interest score table stores values of concept ID, user ID (client terminal ID), TotalZ (user interest score), X, and Y. The definition and calculation method of TotalZ, X, and Y will be described later.

(5-3) Presented content list receiving unit 150
The presented content list receiving unit 150 receives the presented content list as shown in FIG. 7 from the content server 300 and stores it in the content database 160.

(5-4) History information receiving unit 110
FIG. 10 is a flowchart showing a processing procedure and processing contents of the history information receiving unit 110.
(Step S11: History data reception)
The history information receiving unit 110 receives the client terminal ID (or user ID), the list browsing content list, and the detailed browsing content list from the history information transfer unit 350 of the content server 300 via the communication network, and sends the client information to the feature score calculation unit 120. Output.

(5-5) Feature score calculation unit 120
FIG. 11 is a flowchart showing a processing procedure and processing contents of the feature score calculation unit 120. A client terminal ID (or user ID), a list browsing content list, and a detailed browsing content list are input from the history information receiving unit 110 to the feature score calculation unit 120.

(Step S12: Appearance concept extraction)
The feature score calculation unit 120 extracts a concept ID that appears in each content in the detailed browsing content list from the content database 160. Specifically, the “concept ID” associated with each content ID in the detailed browsing content list of FIG. 6 is searched from the content table of the content database 160 of FIG.

  The feature score calculation unit 120 also includes cluster data {cluster ID, list browsing content list, detailed browsing content list}, content ID / concept ID association list {{content ID, {related concept ID, ...}}, ...} and the appearance concept list {concept ID} are generated. Here, the “content ID / concept ID association list” is a list of concept IDs searched based on the content ID. The “appearance concept list” is a list of all concept IDs of concepts that appear in each content included in the list browsing content list and the detailed browsing content list.

(Step S13: Appearance concept extraction)
For each concept ID in the “appearance concept list”, the feature score calculation unit 120 extracts a superordinate concept from the concept system / user interest score database 140 of FIG. 9, and sets the superordinate concept IDs as “appearance concept list” and “content”. To the “ID / concept ID association list”.

  Specifically, the feature score calculation unit 120 searches the concept system table of FIG. 9 to read out the one whose concept ID in the “appearance concept list” matches the “own concept ID”, and the “parent concept ID”. Is extracted. For example, a search is made for a case where the extracted “parent concept ID” matches the “own concept ID” in the concept system table of FIG. 9, and the “parent concept ID” is also extracted as a superordinate concept. Then, the feature score calculation unit 120 associates the concept ID of the superordinate concept with the content ID having the concept ID of the appearance concept that is the source of extraction.

  That is, the extracted “superior concept” is regarded as having been assigned a superordinate concept with respect to “content ID having the original concept ID”, and “appearance concept list” “content ID / concept ID association” Add to list. The extraction of the root concept in the concept system hierarchy is excluded.

(Step S14: Analysis parameter extraction)
The feature score calculation unit 120 calculates the number of appearances for each concept in the “appearance concept list”, extracts analysis parameters necessary for calculating the feature score, and generates an analysis parameter list.
FIG. 12 shows a data configuration example of the analysis parameter list. The analysis parameter list includes, for each cluster ID, the total content S (first total) of the list browsing content list, the total content a (second total) of the detailed browsing content list, and the appearance concept list associated with the cluster ID. N and n are calculated for each concept ID. N (first appearance number) is the number of contents to which the concept ID is assigned in the list browsing content list. n (second appearance number) is the number of contents to which the concept ID is assigned in the detailed browsing content list. Note that N and n are calculated for all concept IDs in the appearance concept list including the superordinate concept added in step S13.

  FIG. 13A shows a schematic diagram of the analysis parameter extraction process. For example, the case where the user browses the details of 10 (= a) contents from a list of 50 (= S) contents is shown. Here, 15 articles (= N) containing the concept of “baseball” among the 50 contents displayed in a list are displayed, and the concept of “baseball” is included in 10 contents viewed by the user. This indicates that there are five (= n) contents including “”.

(Step S15: feature score calculation)
The feature score calculation unit 120 calculates a feature score Z for each concept ID using the analysis parameters S, a, N, and n. FIG. 14 shows details of the feature score calculation process. In FIG. 14, i is a concept identifier, and j is a cluster ID. When H1 (first probability) is the total number S of the list browsing contents included in the list browsing content list and the number N of the contents of the list browsing content where the concept i appears, a detailed browsing content is randomly selected and viewed. In this case, the cumulative probability that the number of detailed browsing contents in which the concept i appears is n or more. When H2 (second probability) is the total number S of the list browsing contents included in the list browsing content list and the number of contents N in which the concept i appears in the list browsing contents, a detailed browsing content is randomly selected and viewed. In this case, the cumulative probability that the number of detailed browsing contents in which the concept i appears is n or less. In the present embodiment, the cumulative probabilities H1 and H2 are obtained by the hypergeometric distribution, but are not limited to this method. Examples of other distributions include a binomial distribution and a normal distribution.

As shown in FIG. 13B, for example, there are five or more contents including the concept of “baseball” among the ten contents viewed by the user using the analysis parameters S, N, a, and n described above. It is shown that the probability of being “0.12”. Here, “0.12” corresponds to the value of the cumulative probability H1.
As an example of using the value of H2, consider the case where n is 0 in the above analysis parameters. In this case, the probability when the number of appearances is 0 or less is calculated. Specifically, in FIG. 13B, the horizontal axis is the value of the item of 0, so “0.02”.

  Then, as shown in FIG. 14, the feature score calculation unit 120 calculates the feature score Z by the inverse function of the standard normal distribution cumulative distribution function using the calculated cumulative probabilities H1 and H2. As shown in FIG. 13C, the feature score Z is obtained by the inverse function of the cumulative distribution function of the standard normal distribution with H1 as the cumulative probability. When H2 is used as the cumulative probability, the feature score Z is obtained with the sign of the return value of the inverse function of the standard normal distribution cumulative distribution function being negative. Using this feature score Z, a concept system update processing unit 130 described later calculates a user interest score TotalZ for the concept of “baseball”.

The feature score calculation unit 120 generates an update target concept list and outputs it to the concept system update processing unit 130. The “update target concept list” is a set of a concept ID, the characteristic score Z calculated above, and a weight w. The concept ID appearing in the update target concept list becomes a node (concept) to be updated in the next concept system update process. The feature score Z and the weight w are calculated for all the concept IDs in the appearance concept list to which the superordinate concept is added. The weight w is a value set for each concept in each cluster ID.
Note that the weight w can be changed as follows when the initial value w = 1 and a user's characteristic operation is performed. For example, with respect to the content presented to the user at the client terminal 200, the user can register as a favorite content, recommend to other users, or input an evaluation for the content.

When the client terminal 200 can transmit an operation history other than the browsing operation to the interest analysis apparatus 100, the following processing is performed.
That is, for example, when a content is registered as a favorite, the feature score calculation unit 120 increases the weight w such that w = 1.5 for all concept IDs included in the content. In addition to the content browsing time, browsing weather, temperature, humidity, season, day of the week, holiday, leisure time, user location information at browsing, scheduler, diary, etc., the value of weight w Can also be changed.

(5-6) Conceptual system update processing unit 130
FIG. 15 shows a processing flow of the concept system update processing unit 130. The concept system update processing unit 130 receives from the feature score calculation unit 120 the client terminal ID (or user ID) and the update target concept list {cluster ID, {concept ID, feature score = Z, weight = w},. Entered.

(Step S16: Concept node value update)
The concept system update processing unit 130 updates the node value of each concept ID in the “update target concept list”. FIG. 16 shows details of the processing of the conceptual system update processing unit 130. The concept system update processing unit 130 uses the user interest score update formula for each concept i shown in FIG. 16 for the concept that appeared in the content (appearance concept) and the concept ID of the superordinate concept of this appearance concept, and the user interest score. TotalZ _in , X _{i (n−1)} , Y _{i (n−1)} values are obtained and input to step S 12 of FIG. 11 corresponding to the cluster ID in the conceptual system / user interest score database 140 of FIG. Each value stored in the user interest score table corresponding to the user ID (client terminal ID) column is updated.

Here, X _{i (n−1)} is the sum of the squares of the weights w of the update target concept list in the past (up to the previous time) for each concept ID (represented by the identifier i here). Similarly, Y _{i (n−1)} is the sum of multiplication of the weight w of the past update target concept list and the feature score Z for each concept ID (represented by identifier i here).

  These X and Y hold intermediate results in the user interest score (TotalZ) calculation process. When giving priority to memory saving / storage, at least three real values of TotalZ, X, and Y are used as variables for each node. It can be realized by holding. When priority is not given to memory saving / storage, all the calculated concept scores and feature scores Z of the respective clusters are stored. In this case, storage of X and Y is not necessary.

In FIG. 16, n is an identifier indicating how many times the concept system update process is performed. A series of processes for obtaining the user interest score TotalZ is performed in units of cluster IDs. When the unit in which this series of processes is performed is counted once, n is the number of times this series of processes has been performed. Is an identifier. i is an identifier of a concept ID. Z _in is a Z value used for each update process of concept i. The Z _ij is a Z value for each of the list browsing content list and the detailed browsing content list, and has a relationship of Z _ij εZ _in . Weight w _in is the weight to be used in each process of updating the concept i. The weight w is the same as that set by the feature score calculation unit 120.

For example, w _in the favorite registration, if the user a special operation other than the inspection of the public, such as with the other user has performed, and (interval to the end from the viewing start) content viewing time, related the degree of content and concepts, The value is changed in accordance with the content browsing time, weather / temperature / humidity / season / day of the week / holiday / leisure at the time of browsing, user location information at the time of browsing, event information collected from a scheduler / diary and the like. It is also changed when specified by other service users and service operators.

Note that forgetting the user interest score, such as to reduce the influence of the history after a certain period of time, attenuates TotalZ, X, and Y when the threshold value of the time interval from the last update to the current time is exceeded. It will be realized.
An example of an attenuation calculation formula is shown. For example, k can be set as an attenuation rate (for example, k = 0.8), and can be calculated as follows.
TotalZ (after attenuation) = k x TotalZ (current)
X (after attenuation) = k ² × X (current)
Y (after attenuation) = k ² × Y (current)
(Step S17: Lower concept node value update)

  Further, the concept system update processing unit 130 extracts the subordinate concept of each concept ID (appearance concept and superordinate concept) in the “update target concept list”, and updates the node value of the subordinate concept. In the extraction of the subordinate concepts, for each concept ID of the “update target concept list”, the concept system table of the concept system / interest degree database 140 shown in FIG. Further, the process of extracting the child concept list with reference to the concept system table for the concept ID of each child concept ID list is repeated.

  The feature score Z used to update the interest level of the lower concept is, for example, the value having the largest absolute value of the feature score among the adjacent parent nodes, the value of the closest higher node, the average of the values of the parent nodes Or a probability-coupled value. In the “update target concept list”, the user interest scores of the concepts updated in step S16 (concepts that appear in the content and higher-level concepts) are not updated.

(5-7) Content evaluation processing unit 170
FIG. 17 is a flowchart showing the processing procedure and processing contents of the content evaluation processing unit 170.
The content evaluation processing unit 170 receives the notification from the content request transfer unit 360 of the content server 300, reads a presentation content list in the format shown in FIG. 7 from the content table of the content database 160, and performs the following content evaluation processing I do.

  That is, content request data as shown in FIG. 20 including the client terminal ID (or user ID) is received from the content request transfer unit 360. In addition, with regard to the above-mentioned presented content list, whether the content registered within the past number of days is to be evaluated based on the prior setting of the service operator or service user (client terminal user) (whether it is included in the presented content list) ) Can be set.

(Step S21: Analysis object concept filtering)
In analysis target concept filtering, when a service operator or a service user designates a concept ID of an analysis target particularly in advance or manually, the concept system / interest degree database 140 is referred to, and the specified analysis target Only concept IDs and subordinate concept IDs are evaluated. The content evaluation processing unit 170 excludes the concept ID held in the input presentation content list from the analysis target according to the conditions set in advance by the service operator or the service user, and generates a “filtered content list”.

  For example, when the user requests a recommendation for content related to baseball, the concept system table shown in FIG. 9 is referred to, and only the subordinate concepts of the concept ID corresponding to baseball are analyzed. The “filtered content list” is a content list obtained by deleting the concept ID associated with each content ID according to the conditions set in advance by the service operator or the user by the above processing. The “filtered content list” has the same data configuration as the presented content list of FIG.

(Step S22: Score evaluation)
The content evaluation processing unit 170 calculates the evaluation score of the content included in the “filtered content list” and generates a content score list as shown in FIG. The content score list has a content ID, an evaluation score, a content body, and a content registration time.

FIG. 19 shows an example of a method for calculating the evaluation score. For example, according to the content evaluation formula shown in FIG. 19, the evaluation score EntityZ _x for the content x is the user interest score TotalZ _{i of} the concept i, the relevance w _i between the content x and the concept _i (or the importance of the concept i), and It can be calculated using a set p of concept IDs appearing in the content x. The concept identifier i corresponds to the concept ID in the set p.

The user interest score (TotalZ) used in the calculation of FIG. 19 is the client terminal ID (or user) from the user interest score table (FIG. 9) of the concept system / user interest score database 140 for the concept ID related to each content. Read out based on (ID). In FIG. 19, when the content 1 in which the concept K, the concept B, and the concept D appear is the evaluation content, the evaluation score EntityZ _x = 0.18 is calculated using the TotalZ, w of the concept K, the concept B, and the concept D. it can. On the other hand, when the content 2 in which only the concept B appears is the evaluation content, the evaluation score EntityZ _x = −0.3 can be calculated using the TotalZ, w of the concept B. The content 1 having a large evaluation score EntityZ _x is displayed with priority.

In addition, the evaluation score EntityZ _x can be obtained by the method of Modifications 1 to 3 below.
As a first modification, it is obtained by EntityZ _x = MAX (TotalZ _i * w _i ). MAX (TotalZ _i * w _i ) is a function that returns the maximum value of TotalZ _i * w _i for i∈p.

The second modification, the value of EntityZ _x When the value of the MAX (TotalZ _i * w _i) exceeds the threshold value, the return value of _{MAX (TotalZ i * w i)} . MAX (TotalZ _i * w _i ) is a function that returns the maximum value of TotalZ _i * w _i with i∈p. It does not exceed the threshold, and EntityZ _x the resulting content evaluation formula of Figure 19. MAX () is a function that returns the maximum value with TotalZ _i * w _i of i∈p. The threshold is a value set by the service operator.

As a third modification example, only a value of iεp having a positive TotalZ _i is extracted, and a value integrated by the content evaluation formula of FIG. 19 is defined as EntityZ _x .
(Step S23: Sort contents in score order)
The content evaluation processing unit 170 sorts the content included in the content score list in descending order of the evaluation score EntityZ _x and outputs the sorted content score list to the sorted content score list transmission unit 180.

(5-8) Sorted content score list transmission unit 180
The sorted content score list transmission unit 180 transmits the sorted content score list and client terminal ID (or user ID) input from the content evaluation processing unit 170 to the content server 300 via the communication network.

(5-9) Attribute Interest Score Update Unit 195
The attribute interest score update unit 195 creates a model of the attribute for each user attribute using a model of another user that matches the attribute.
FIG. 21 shows an outline of processing for creating a model related to an attribute by the interest analysis apparatus 100, and a case where a model of attribute A is created using models of user B and user C that match the attribute A is shown. This is just an example. The attribute interest score update activation unit 190 periodically activates processing in a time zone where the processing load of the interest analysis device 100 is low, such as a midnight time zone. When the process is started, the attribute interest score update unit 195 first identifies a user group that matches the attribute A, for example, the user B and the user C, using the attribute information stored in the user / attribute information database 198. Next, X, Y, and TotalZ of the attribute A are calculated using the formula shown in FIG. 32 based on the identified X and Y of the user. Then, the calculation result of X, Y, and TotalZ of the attribute A is stored in the attribute interest score database 196.

  FIG. 22 shows an overview of processing when the interest analysis apparatus 100 creates a higher-level attribute model using a lower-level attribute model. The attribute A model is represented by a child attribute of the attribute A. The case where it creates using the model of a certain attribute B and attribute C is illustrated. As described above, the attribute interest score update activation unit 190 periodically activates processing in a time zone where the processing load of the apparatus is low, such as a midnight time zone. When the process is started, the attribute interest score update unit 195 specifies a child attribute group of the attribute A, for example, the attribute B and the attribute C, using the attribute information stored in the user / attribute information database 198. Next, X, Y, and TotalZ of attribute A are calculated using the formula shown in FIG. 34 based on the specified attributes X and Y. Then, the calculation result of X, Y, and TotalZ of the attribute A is stored in the attribute interest score database 196.

FIG. 23 shows an outline of processing related to creation of an initial model of a new user (initial use user) using an attribute model by the interest analysis apparatus 100.
In the figure, a new user inputs his / her own attribute and a ratio W for probabilistically connecting each attribute in his / her client terminal 200. Then, the attribute presentation unit 250 of the client terminal 200 transfers the input information representing the user's own attribute and the probability coupling ratio W to the interest analysis device 100 via the attribute transfer unit 370 of the content server 300. .

  In the interest analysis apparatus 100, the initial user interest score creation unit 197 reads an attribute model that matches the transferred attribute from the attribute interest score database 196. Then, the initial interest score TotalZ for the initial use user is calculated by stochastic combination using the equation shown in FIG. 29, and the calculated initial interest score TotalZ is used as the initial model of the initial use user as a conceptual system / user interest. Store in the score database 140.

(6) Effects of Embodiments As described above in detail, according to one embodiment, the following functions and effects can be achieved.
(1) By defining a list that can be selected by the user and analyzing the number of occurrences of the concept in content selection from that list, the rareness of each concept can be considered and the concept not selected from the list Since the history feature can be used, it is possible to estimate the user's interest with high accuracy.

  (2) Further, as shown in steps S15 and S16 above, when calculating the feature score and the steer user interest score, the situation of the user at the time of browsing, the characteristics of the browsing operation (favorite registration, long-time browsing, etc.), etc. Can be reflected via the weighting coefficient (weight w), so that the user interest score can be obtained with higher accuracy.

(3) In addition, in the content browsing history analysis where the concept defined by taxonomy (ontology) etc. was added as a meta tag, it was difficult to rationally reflect the rareness of concept appearance in the analysis. It was necessary to make adjustments on the ontology structure side, such as unifying the abstraction level of the concept to be given to the content by fixing the depth of the structure. However, in this embodiment, it is possible to analyze by integrating the content given the higher concept by the update process of the user interest score by the ontology structure considering the rareness of concept appearance and the browsing history of the content given the lower concept Become. For this reason, it is possible to reduce the restrictions on the taxonomy (ontology) used for the analysis and to reduce the maintenance, operation, and management costs of the taxonomy (ontology).
(4) Furthermore, by calculating the user's evaluation score for content using the user interest score, it is possible to recommend content that matches the user's interest.

  (5) In addition, a model related to an attribute is created using a model of an existing user, and a model that matches the own attribute presented by the new user (initial use user) is used as the initial model of the new user. For this reason, even when the user uses the information recommendation service for the first time, a relatively good recommendation result can be obtained. In addition, since the model relating to the attribute is created by stochastic coupling using a model of an existing user, a model for each attribute can be created with a small amount of calculation.

(6) Furthermore, since the probability coupling process is performed by selecting a midnight time zone where the processing load of the apparatus is low, there is an advantage that it can be performed without increasing the processing load of the apparatus.
(7) Furthermore, when an attribute has a hierarchical structure, an attribute model having a hierarchical structure can be created with a small amount of calculation by creating a higher attribute model using a lower attribute model. .

[Other Embodiments]
The present invention is not limited to the above embodiment. For example, the configuration of the interest analysis apparatus, the model creation process for each attribute and the processing procedure and processing contents of the initial model creation process, the type of attribute, etc. Various modifications can be made without departing from the scope of the invention.

  In short, the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, you may combine suitably the component covering different embodiment.

  DESCRIPTION OF SYMBOLS 100 ... Interest analysis apparatus, 200 ... Client terminal, 300 ... Content server, 110 ... History information receiving part, 120 ... Feature score calculation part, 130 ... Concept system update processing part, 140 ... Concept system / user interest score database, 150 ... Presented content list receiving unit, 160 ... content database, 170 ... content evaluation processing unit, 180 ... sorted content score list transmitting unit, 190 ... attribute interest score update starting unit, 195 ... attribute interest score updating unit, 196 ... attribute interest score Database, 197 ... initial user interest score creation section, 198 ... user / attribute information database, 210 ... history collection section, 220 ... history information transmission section, 230 ... content presentation section, 240 ... content request transmission section, 250 ... attribute presentation section 310 ... Content transmission processing , 320 ... it sorted provided content list receiving section, 330 ... provided content list transmitting section, 340 ... provided content list input unit, 350 ... History information transfer section, 360 ... content request transfer unit, 370 ... attribute transfer unit.

Claims (7)

The computer uses the information representing the scarcity of concept appearances to estimate the interests of information users, and uses the estimation results to create a model that represents the degree of interest in a given concept for each information user. A method for analyzing the interest of each information user using a created model,
A first step of creating a model for each attribute of the existing information user based on information representing a content selection history by the information user and storing the model in a first storage unit;
A second process of acquiring attribute information of a new information user;
Based on the acquired attribute information, an existing information user attribute model that matches the attribute of the new information user is selectively read out from the first storage unit, and the read model is A model that represents the degree of interest in the predetermined concept of the new information user is created, and the created model is stored in a second storage unit as an initial model of the new information user. 3 process,
And a fourth step of analyzing the interest of the new information user using the model stored in the second storage unit as an initial model.   In the case where the attribute of the existing information user has a first hierarchy corresponding to a lower concept and a second hierarchy corresponding to a concept higher than the first hierarchy, the first process includes: A model of a subordinate concept corresponding to the first hierarchy is created, and a model of a superordinate concept corresponding to the second hierarchy is created based on the model of the subordinate concept created for the first hierarchy. The interest analysis method according to claim 1.   The first step sets a threshold value or a time zone in which the processing load of the computer is equal to or less than a preset value, and when the processing load becomes equal to or less than the set threshold value or the set time zone The interest analysis method according to claim 1 or 2, wherein a process of creating a model for each attribute is executed when the value becomes. The information that represents the scarcity of concept appearance is used to estimate the interest of the information user, and the estimation result is used to create a model that represents the degree of interest in a given concept for each information user. An interest analysis device that analyzes the interest of each information user using a model,
Means for creating a model for each attribute of the existing information user based on the information representing the content selection history by the information user;
First storage means for storing the created model for each attribute of the existing information user;
Means for acquiring attribute information of a new information user;
Based on the acquired attribute information, a model of an existing information user attribute that matches the attribute of the new information user is selectively read out from the first storage means, and the read model is Means for creating a model that represents the degree of interest in the predetermined concept of the new information user based on;
Second storage means for storing the created model relating to the new information user as an initial model of the new information user;
An interest analysis apparatus comprising: means for analyzing the interest of the new information user based on the initial model stored in the second storage means.   The means for creating a model for each attribute has a first hierarchy in which the attributes of the existing information user correspond to a lower concept and a second hierarchy corresponding to a concept higher than the first hierarchy. Then, a lower concept model corresponding to the first hierarchy is created, and a higher concept model corresponding to the second hierarchy is created based on the lower concept model created for the first hierarchy. The interest analysis apparatus according to claim 4, wherein:   When the processing load of the interest analysis device is set to a threshold value or a time zone where the processing load is equal to or less than a preset value, and when the processing load is equal to or less than the set threshold value or when the set time zone is reached 6. The interest analysis apparatus according to claim 4, further comprising means for causing the means for creating a model for each attribute to execute a process for creating the model for each attribute.   The program which makes the said computer perform the process corresponding to the process with which the interest analysis method in any one of Claims 1 thru | or 3 is provided.

Images